NTU AI Ruler Predicts Atrial Fibrillation Stroke Risk with 90% Accuracy
A research team from National Taiwan University has developed an AI tool, dubbed the 'AI ruler,' that predicts stroke risk from atrial fibrillation with 90% accuracy. This innovation helps clinicians precisely determine who needs anticoagulant medication, reducing unnecessary bleeding risks. The findings were published in 'npj Digital Medicine,' marking a significant step towards personalized precision medicine.
📋 Article Processing Timeline
- 📰 Published: April 10, 2026 at 13:54
- 🔍 Collected: April 10, 2026 at 14:00 (6 min after Published)
- 🤖 AI Analyzed: April 15, 2026 at 21:44 (127h 44m after Collected)
At a press conference this morning, Dr. Lai Chao-lun, Director of Internal Medicine at Hsinchu NTU Hospital, presented that anticoagulants are commonly used clinically to prevent stroke, but these drugs may increase the risk of bleeding, ranging from minor issues like gum bleeding and hemorrhoids to severe crises such as gastrointestinal bleeding and cerebral hemorrhage.
In the past, the risk for 'intermediate' groups was often overestimated, leading some patients to take medication unnecessarily, thereby increasing their bleeding risk. Dr. Lai stated that the strength of the new AI lies in its ability to function in these 'ambiguous' risk groups, clearly delineating risk boundaries so that those who need medication don't miss out, and those who don't need it suffer less, achieving 'precision medicine when it's needed.'
The NTU Hospital research team developed the model using 9511 newly diagnosed atrial fibrillation cases from the NTU Hospital Integrated Data Center between 2007 and 2016. Through a dual-model design, they balanced prediction accuracy with model transparency, preventing the AI from becoming an unexplainable 'black box.'
The research team further applied the model to cases at Hsinchu NTU Hospital (1300 cases) and Yunlin Branch (1242 cases) for validation. The results confirmed that the model has extremely high applicability and stability across different clinical settings, demonstrating its potential for cross-hospital application.
In addition, the study incorporated explainable AI techniques, clearly presenting the direction and weight of influence of various risk factors. Physicians can not only obtain risk prediction values but also understand the underlying reasons, which helps with doctor-patient communication and clinical decision-making.
Dr. Lai compared traditional assessment tools to a 'hard ruler,' which only considers age, gender, and past medical history, with an accuracy of about 60%. The new model is like a 'flexible tape measure,' incorporating diverse information such as lung disease, liver disease, and medication history, significantly increasing accuracy to nearly 90%.
This research breaks through the limitations of traditional clinical scoring tools, and the results were officially published on April 7th this year in 'npj Digital Medicine,' ranked first in the field of digital healthcare.
It is estimated that about 150,000 to 200,000 adults over 35 in Taiwan are exposed to the risk of atrial fibrillation. Dr. Lai stated that atrial fibrillation patients have a 5 to 10 times higher risk of stroke than the general population, but anticoagulants carry about a 2% annual bleeding risk. If a patient's stroke risk is less than 2%, taking medication is more dangerous and may not be worthwhile. This is where AI can help with precise calculations.
The NTU team explained that this architecture emphasizes 'explainability,' allowing doctors to see the logic within the AI's 'brain,' thereby creating personalized medical decisions. It is currently in the retrospective research stage and awaits further clinical validation. Dr. Lai stated that this is just the beginning, and they hope to implement more medical AI in the future to truly achieve health equity. (Edited by Chen Ching-fang) 1150410
In the past, the risk for 'intermediate' groups was often overestimated, leading some patients to take medication unnecessarily, thereby increasing their bleeding risk. Dr. Lai stated that the strength of the new AI lies in its ability to function in these 'ambiguous' risk groups, clearly delineating risk boundaries so that those who need medication don't miss out, and those who don't need it suffer less, achieving 'precision medicine when it's needed.'
The NTU Hospital research team developed the model using 9511 newly diagnosed atrial fibrillation cases from the NTU Hospital Integrated Data Center between 2007 and 2016. Through a dual-model design, they balanced prediction accuracy with model transparency, preventing the AI from becoming an unexplainable 'black box.'
The research team further applied the model to cases at Hsinchu NTU Hospital (1300 cases) and Yunlin Branch (1242 cases) for validation. The results confirmed that the model has extremely high applicability and stability across different clinical settings, demonstrating its potential for cross-hospital application.
In addition, the study incorporated explainable AI techniques, clearly presenting the direction and weight of influence of various risk factors. Physicians can not only obtain risk prediction values but also understand the underlying reasons, which helps with doctor-patient communication and clinical decision-making.
Dr. Lai compared traditional assessment tools to a 'hard ruler,' which only considers age, gender, and past medical history, with an accuracy of about 60%. The new model is like a 'flexible tape measure,' incorporating diverse information such as lung disease, liver disease, and medication history, significantly increasing accuracy to nearly 90%.
This research breaks through the limitations of traditional clinical scoring tools, and the results were officially published on April 7th this year in 'npj Digital Medicine,' ranked first in the field of digital healthcare.
It is estimated that about 150,000 to 200,000 adults over 35 in Taiwan are exposed to the risk of atrial fibrillation. Dr. Lai stated that atrial fibrillation patients have a 5 to 10 times higher risk of stroke than the general population, but anticoagulants carry about a 2% annual bleeding risk. If a patient's stroke risk is less than 2%, taking medication is more dangerous and may not be worthwhile. This is where AI can help with precise calculations.
The NTU team explained that this architecture emphasizes 'explainability,' allowing doctors to see the logic within the AI's 'brain,' thereby creating personalized medical decisions. It is currently in the retrospective research stage and awaits further clinical validation. Dr. Lai stated that this is just the beginning, and they hope to implement more medical AI in the future to truly achieve health equity. (Edited by Chen Ching-fang) 1150410